openghg_inversions.models.pollution_event#

Pollution-event-scaled observation-error construction.

For an observation Y with reported standard deviation error, the historical model uses

epsilon = max(sqrt(error**2 + (pollution_event * sigma)**power), min_error).

The pollution event is either the modelled pollution contribution or the observation after removing an explicitly supplied modelled baseline. Fixed aggregation covariance is an optional, explicitly selected addition.

This module owns the pollution-event PyMC graph and therefore requires an active model context. Its PEFO-only pollution and baseline inputs do not cross RHIME’s general likelihood-builder seam.

openghg_inversions.models.pollution_event.add_pollution_event_likelihood(*, observations: DataArray, observation_error: DataArray, minimum_error: DataArray, aggregation_error: AggregationError, mean: TensorVariable, pollution_mean: TensorVariable, pollution_event_baseline: TensorVariable | None, sigma_alignment: SigmaAlignment | None, sigma_prior: Mapping[str, Any], power: Mapping[str, Any] | float, pollution_events_from_obs: bool, no_model_error: bool, retain_unused_sigma: bool = False, output_dim: str = 'nmeasure') → TensorVariable#

Build RHIME’s pollution-event-scaled Gaussian likelihood.

Parameters:

  • observations – Observed mole fractions.

  • observation_error – Reported observation-error standard deviations.

  • minimum_error – Minimum total-error standard deviations.

  • aggregation_error – Validated fixed aggregation-error representation.

  • mean – Completed modelled concentration, including the full baseline.

  • pollution_mean – Modelled pollution contribution used for mismatch scaling when pollution_events_from_obs is false.

  • pollution_event_baseline – Modelled baseline removed from observations when pollution_events_from_obs is true.

  • sigma_alignment – Observation alignment for mismatch parameters when model error is enabled or unused sigma is retained.

  • sigma_prior – Prior specification for mismatch parameters.

  • power – Exponent or prior specification used in mismatch scaling.

  • pollution_events_from_obs – Whether observed rather than modelled pollution enhancements control mismatch scaling.

  • no_model_error – Whether to omit inferred mismatch error.

  • retain_unused_sigma – Whether to retain the historical disconnected sigma variable when model error is disabled.

  • output_dim – Observation dimension name.

Returns:

Observed Gaussian variable named y. The component also creates the canonical total-error variable epsilon.